Uğur Arıkan

Have been working on and driving various Operations Research projects such as:

• Human-centric last-mile delivery with route generation algorithms utilizing OR, statistics and Machine Learning to combine valuable tacit knowledge of experienced couriers with optimization algorithms. This project has been selected as a finalist for Wagner Prize in 2022.
• Revenue maximization project for air capacity sales with large scale global network models.
• Scenario generation, optimization and analysis tool to support global air network design decisions.
• Container loading algorithms for trucks and air cargo with the objectives such as to minimize cost, increase utilization and minimize delays.
• Asynchronous optimization api for long running optimization problems to make it convenient and efficient to build, deploy and test optimization services for tactical and strategical decision making projects.
• Generic and efficient multi commodity network flow solver with flexible side constraints and adaptable solution approaches to support various network design projects.
• Generic and efficient traveling salesperson problem solver with flexible side constraints such as time windows or LIFO to be used as the core solver in various vehicle routing problems.

• I had the opportunity to work closely with my supervisors Prof. Karthik Natarajan and Prof. Selin Damla Ahipaşaoğlu. Our research focused on discrete choice models and stochastic traffic equilibrium. We worked on to make use of robust optimization techniques to define and model the distributionally robust traffic equilibrium as a new convex optimization formulation which is provably efficiently solvable and has the flexibility of allowing for general marginal distributions. Having our research focus on transportation systems, I participated the MIT Transportation Networks and Smart Mobility: Methods and Solutions program lead by Prof. M. Ben-Akiva.
• I have been a lecturer of the Capstone course for four years where I taught and supervised real-life projects of senior students. I was very keen on this task particularly due to the multidisciplinary nature of the projects and collaboration both with students and industry.
• I feel very lucky for the opportunity to work with Prof. Peter Jackson on Arcadia with Capella. Systems Design is always a favorite topic for me. Additionally, working with Peter Jackson was pure fun. You may find here the tutorial we created.

• I had been a teaching assistant for various courses such as Operations Research, Production & Service Information Systems, Stochastic Models, Revenue Management, Systems and Organizations.
• I was a part of the Systems Design committee with supervising and organizational tasks for the real-life systems design projects of the senior students.
• I took place in industrial OR projects such as the assembly line design project for car manufacturer Tofaş that simultaneously optimizes make-or-buy / outsourcing decisions of sub-components.

During my graduate studies, I had taken place in various software projects at my university mostly focusing on process optimization for in-campus processes.

Title of Dissertation: Airline Disruption Management

Supervisors: Prof. Sinan Gürel and Prof. M. Selim Aktürk

Air transportation has been a very important sector due to its operational and economical aspects. In this growing industry, airlines try to analyze a very large and complex network in order to plan their operations. However, operations cannot be operated as scheduled due to irregularities in operations, namely disruptions. Original schedules of aircraft, crew members and passengers become infeasible; and airlines try to recover their schedules with minimum disruption and recovery costs. Moreover, due to the dynamic nature of the operations, operation controllers need to take action in real time; which makes the rescheduling efforts more challenging. more The traditional approach is to recover the schedules of aircraft, crew members and passengers sequentially. The reports show that the sequential approach results in high recovery costs and passenger inconvenience. In the first part of our study, we integrate passenger recovery and cruise speed control with aircraft rescheduling. A special emphasis is placed on passenger recovery. In our experiments, we observed that integration of cruise speed control option helps mitigate delays and create new swap opportunities, which is one of the most common recovery actions in practical disruption management. The problem is initially formulated as a mixed integer nonlinear programming (MINLP) model. We reformulate the problem as a conic quadratic mixed integer programming (CQMIP) problem. We were able to solve about 93% of the instances of a large airline network to optimality within 60 seconds. In the second part, we propose a new general network representation of the problem that can integrate any entity that is transported through the flight network of the airline. We show that the proposed representation is more compact than the traditional time-space network and flight string representations. The flexibility of the representation allows to integrate all possible recovery actions together with the cruise speed control which is studied by a few authors only. We investigated and created disruption scenarios with varying severities using the real data of a major airline company. Moreover, we propose new realistic passenger delay cost formulations. Finally, we propose a CQMIP formulation of the integrated airline recovery problem using the proposed representation. Proposed approach was able to find the best recovery actions of a major U.S. airline within reasonable solution times. In the final part, using the flexibility of the proposed representation, we propose a heuristic approach to provide fast and good solutions to very large airline networks. The heuristic utilizes the advantages of rescheduling and relies on the fact that the original schedules are optimal provided that there are no disruptions. The algorithm uses the interdependencies among the recovery actions defined by the proposed networks to quickly reduce the problem size and find a subset of eligible recovery actions. The heuristic aims to find the best reduced feasible region that provides a solution within the predetermined time limit. For large instances with severe disruptions that are optimized within around two hours, proposed approach provides solutions with about 4.69% more costs within five minutes. less

Title of Dissertation: Two-Sided Assembly Line Balancing: Models and Heuristics

Supervisors: Prof. Ömer Kırca

A two-sided assembly line is a special type of assembly lines which is mainly used in production of large products such as trucks and buses. Two-sided assembly line balancing problem (TSALBP) is relatively a recent type of assembly line balancing problems. In addition to common constraints of assembly lines, a subset of the tasks in TSALBP also have side constraints. more In the first part of our study, we develop a mixed-integer programming (MIP) formulation which uses integer station assignment variables instead of binary variables as used in most proposed formulations in the literature. We were able to provide the very first optimal solutions provided by an exact approach to some benchmark problems. Due to the encouraging results of the mathematical formulation, we propose a rolling-horizon heuristic integrated with the formulation in the second part. The heuristic proceeds on the assembly line solving smaller MIP problems using parameters to control the size of the subproblems and to decide on accepting or rejecting the solutions. The heuristic managed to optimize all benchmark problems within 10800 seconds (a commonly used time limit on the benchmark problems). In the third part, we proposed a second heuristic approach to provide fast solutions. We have been inspired by the multiple-rule heuristic of Boctor (1995). The proposed heuristic is an enhanced version in the sense that it uses a greater number of logical rules and their combinations; and the rules are designed to handle side constraints. The polynomial time heuristic was able to achieve optimal solutions to the largest data sets within 30 seconds. less